On Wed, Dec 05, 2001 at 03:28:06PM +0100, V. T. Mueller wrote:
> On Wed, 5 Dec 2001, Ollie Cook wrote:
> > mysql_servers = set1/host1::3306/database1/user1/password1\
> > set1/backup-host1::3306/database1/user1/password1\
> > set2/host2::3306/database2/user2/password2\
> > set2/backup-host2::3306/database2/user2/password2
> > This would still allow for redundancy, but still cut down on the
> > erroneous connections to database which have no hope of answering
> > the query.
>
> Hm. I could live with that, along with a certain amount of pain.
>
> Again: If you encounter performance problems on host1, put in another cpu.
> If you encounter disk space problems, put in disks and increase the
> respective logical volume [or deal with disks, mountpoints and syslinks if
> your OS is not lvm capable]. At least it appears as if triggering a little
> SQL error msg does worry you you have a rather high volume site - so you
> should be able to invest in hardware instead of tweaking software in a
> doubtful direction.
>
> The software's job, however is to be as simple as possible.
> Experience tells us that software / site design changes provoke failures.
> But I've never seen a service failing because someone added diskspace ;)
>
> Maybe my view on this is too much HA-coloured, but at least this is what I
> think.
>
> Someone else's opinion?
Mine, again, I'm afraid! :)
We're digressing slightly, but I don't agree with your view that you
should just throw money at a problem rather than make an attempt to
fix it, or in this case, get it fixed.
Exim's current behaviour, for historical reasons I imagine, is wasteful
in terms of network bandwidth, in terms of CPU time (on the mail server
and the database server) - and increases delays to mail delivery which
aren't necessary. All I am suggesting is an improvement in this mechanism,
whereby you can direct queries to the database server(s) that you know will
be able to answer the query.
I understand you objection to the first suggestion I made, but the
second one gives you the same redundancy, without the waste. I may be
missing what you're getting at though, so could you explain it a
different way?
In our experience, software changes only provoke failures without
thorough testing. Philip is always very thorough when it comes to this
kind of thing and we've never found a problem testing new features in Exim.
In the past, Philip has taken this kind of constructive criticism well,
and if he has seen fit has made the appropriate amendments to the code.
I guess the last laugh is with Dr Hazel on this one! :)
O.
--
Oliver Cook Systems Administrator, ClaraNET
ollie@??? 020 7903 3065
